Above Consensus

Stocks and supply chains. Finding what's actually scarce before most, and what the street has wrong. Positions disclosed. Not investment advice.

[Above Consensus @vsconsensus]

SK Hynix printed a 76% operating margin. Revenue up 257% YoY. HBM4 in mass production with yields near mature HBM3E.

The stock fell.

What the tape sold was the capex raise to the high 40T won range. What it ignored was in the same call: five-year LTAs with roughly ten customers, structured with deposits.

Every prior memory cycle broke because capacity got built into forecast demand. This capex is gated against contracted demand with cash down.

That's the difference, and it's the whole thesis. If the LTAs hold, 2028 supply doesn't break pricing the way 2018 did.

Long memory.

[Above Consensus @vsconsensus]

@firstadopter Congrats. One of the few greats on here.

[Above Consensus @vsconsensus]

Teng Yan (@tengyanAI) made the $CLF case in June: the sole U.S. producer of grain oriented electrical steel, a 50% tariff wall, DPA designation, and stainless and electrical steels combined at just 3% of Q1 shipment volume.

His conclusion: a steel stock with a mispriced electrical steel option.

I agree with the setup. I think the option is narrower than it looks, and the reason is which GOES.

The bull case runs through qualification. Transformer OEMs are adding U.S. capacity on 2027-2028 timelines and will need qualified domestic steel. But domestic supply only matters if it can meet the required specifications.

Cliffs produces high permeability and domain refined grades, but its GOES range tops out at 920 mm wide, below what some of the largest transformer cores require. Its higher end capacity is also limited relative to the full range of material demanded by transformer manufacturers.
That helps explain why most GOES used in U.S. large power transformer production is still imported. This isn't simply a tariff avoidance choice. In parts of the market, there is no qualified domestic substitute to switch to.

Tariffs protect existing production. They don't create wider coils, higher grade capacity, or compress a qualification cycle from years into quarters.

So the option is real, but narrower and slower moving than the headline suggests. It is most valuable where Cliffs' existing grades and widths already meet OEM specifications, and weakest at the highest performance end, where the data center driven grid buildout is most constrained.
That leaves pricing power with foreign high end producers wherever no domestic substitute is qualified. The tariff raises the price without shifting the share.

$CLF owns the domestic monopoly. It doesn't yet own the high-end bottleneck.

No position.

[Above Consensus @vsconsensus]

@aleabitoreddit Nvidia just standardized Isaac GR00T on Unitree's H2 chassis.

Also worth noting the NDAA version covers military procurement only. The commercial ban is the GUARD Act, still a bill.

Removing the competitor doesn't build the supply chain. Same mistake as the drone ban.

[Above Consensus @vsconsensus]

Kalshi and Polymarket: 80% no change tomorrow. CME FedWatch: 71.7%.

Eight points apart on the same event.

Rates desks are hedging Citadel's call. Prediction markets aren't. I'd take the prediction market side.

No hike.

[Above Consensus @vsconsensus]

@WisemanCap The uncertainty is 5 Chinese DUV tools this year, 20 in 2027. ASML shipped 131 immersion units last year. That’s not froth coming out, that’s a headline getting mispriced.

[Above Consensus @vsconsensus]

@DeItaone The domestic tool is around 2007 era ASML capability at 5 units a year. It’s not competitive on performance and it doesn’t touch leading edge.

[Above Consensus @vsconsensus]

GPU lead times are measured in months. Large power transformer lead times run 48 to 60 months.

The constraint moved downstream. The market is still fixated on the chip.

Global gas turbine orders hit 110 GW against 60-70 GW of manufacturing capacity. GE Vernova is nearly sold out through 2030, with deliveries stretching into 2031. Turbine pricing is up 195% versus 2019.

But the chokepoint goes one layer deeper. Transformers need grain oriented electrical steel, and five companies control roughly 3.5-4M tons of global capacity. That's a steel mill problem, and steel mills don't scale on a data center timeline.

Compute can be ordered in months. The grid to power it takes years.

$GEV is the obvious way to play this. The interesting names are upstream, in electrical steel and the transformer supply chain. Posting that breakdown this week.

No position.

[Above Consensus @vsconsensus]

@TradexWhisperer CPO isn’t a 2029 technology. Broadcom is shipping its third gen Tomahawk CPO platform and NVIDIA’s Quantum X Photonics is in customers’ hands this year. 2029 is when CPO reaches broad volume, not when it starts. If anything gets skipped, it’s the NPO middle step.

[Above Consensus @vsconsensus]

@firstadopter Real memory scarcity, priced through a 6.73% float

[Above Consensus @vsconsensus]

@FT CXMT has 7.67% share. Micron has 3x that at half the multiple

[Above Consensus @vsconsensus]

@aleabitoreddit The $350B chip order is revenue. The $250B guarantee is a contingent liability that pays out precisely when the chip order doesn't. Capex beneficiaries win either way

[Above Consensus @vsconsensus]

@jukan05 You’re conflating strategic risk with financial impact. NVIDIA already assumes essentially 0 China AI GPU revenue. If China builds a competitive stack, that’s more lt ecosystem risk, not a hit to today’s business

[Above Consensus @vsconsensus]

@zephyr_z9 The bear case was basically: use 4× the chips, more power, more networking, and then comp it to 2 year old NVDA hardware. That’s pretty silly

[Above Consensus @vsconsensus]

@KyleSamani Nothing worse than a guy who missed the entire trade and suddenly became an expert

[Above Consensus @vsconsensus]

@TiffanyFongEtc It’s normal escheatment laws. All 50 states have them, some have applied this to crypto already (not just California)

[Above Consensus @vsconsensus]

@HeroDividend $4 trillion spending $1 billion would be proportional to someone with $1m spending $250. Probably sustainable for a while

[Above Consensus @vsconsensus]

@JSeyff This was the weakest sentence of any related crime, removing Madoff, Tom Petters 7B fraud got him 110 years, Scott Rothstein's 1.2B got him 50

[Above Consensus @vsconsensus]

@EconomyApp Your Glassdoor research made you miss a 15x and you’re here touting it as a good investment resource